An AI chatbot for your website is worth buying ready-made when you need answers from your own pages and files, a hand-off to staff, and the common integrations; a monthly plan covers it. Build a custom one only when the bot must follow rules or reach systems no product supports. Check a free trial before you pay, and judge it on your own questions.
An AI chatbot for a website is a chat widget that reads your own content, such as help articles and product pages, and answers visitors in their own words. It looks up the relevant passages at the moment someone asks, rather than being retrained on them, so its answers are only as good as the content behind it. The technical name for that lookup step is retrieval-augmented generation, or RAG.
What does an AI chatbot for a website actually do?
It takes a question, finds the matching passages in your content, and writes an answer from them, with a link to the source and a way to reach a person. The model is the small part; your content and the rules around it do most of the work.
- Retrieval. Your pages, PDFs and FAQs are split into chunks and indexed. Each question pulls the closest chunks and sends them to the model.
- An answer with its source. A good bot shows the page it used, so a visitor can check it and you can find the article to fix.
- A way out. When nothing relevant is found, it should say it does not know and offer a person, not guess.
- Optional account lookups. For a signed-in customer it can read their own order or invoice, with identity taken from the login session and never from an email typed into the chat.
Anything that changes a record, such as a refund or a cancellation, is a different and riskier job. The AI chatbot vs AI agent guide covers where that line sits.
Should you buy a ready-made tool or build a custom chatbot?
Buy ready-made for the common needs, and build only for the uncommon ones. Most website chatbots do the same handful of jobs, and a hosted tool does them on a monthly plan without a build.
| Your situation | Start with | Why |
|---|---|---|
| Visitors ask things your pages and files already answer | A ready-made tool | A monthly plan covers retrieval, a widget and hand-off without a build. |
| You want it live this week and will test as you go | A ready-made tool on a free trial | You watch it answer your own questions before paying. |
| Answers depend on a signed-in customer's own account | Either, with read-only lookups | A product with your platform's connector, or a custom read-only link. Identity comes from the login. |
| The bot must follow rules no settings page exposes, or live inside your own app | A custom build | Only code can enforce logic a product does not offer. |
| It must refund, cancel or change records on its own | A person, or an agent behind approval | A wrong action costs more than a wrong answer. |
| Your content is thin or out of date | Fix the content first | Every tool repeats what you give it. |
If a wrong answer would cost little and an easy correction fixes it, a ready-made tool is usually enough. If a wrong action would cost real money, keep a person in the loop whichever route you choose.
What do ready-made AI chatbot tools cost today?
Entry plans run from about $10 to $40 a month, but the billing models differ so much that the headline price tells you little on its own. Below are four tools with their entry prices as I read each vendor's pricing page on 20 October 2026.
| Tool | Entry price (20 Oct 2026) | What the entry plan covers | How AI usage is billed | Free trial or refund |
|---|---|---|---|---|
| MagizAI (my product) | $10/mo BYOK, excludes GST | Unlimited messages on your own AI key, 3 bots and 3 seats, WhatsApp and WHMCS connectors | You pay your AI provider directly | No free tier; full refund within 7 days of first payment |
| Chatbase | $40/mo Hobby ($32/mo billed yearly) | 700 message credits a month, 1 agent, 40 MB of training content | Included in the plan's credits | 7-day trial |
| Tidio (Lyro AI) | Lyro from $32.50/mo | From 50 Lyro AI conversations a month, website scraper and FAQ import | Included; the Premium tier bills per resolution | Free plan with the first 50 Lyro conversations free for life; 7-day trial |
| Intercom (Fin AI Agent) | $0.99 per Fin outcome | Pay per outcome Fin produces; helpdesk seats from $29 each a month | Per outcome, charged once per conversation | 14-day trial |
MagizAI is my own product and I am its founder, so I gain if you choose it; the figures above come from magiz.ai/pricing, and I describe how it reads WHMCS on the MagizAI page. The others are from Chatbase, Tidio and Intercom, read the same day. Tidio says its customers resolve about 75% of conversations with Lyro on average, a vendor figure I cannot independently check.
What do the plan limits really mean?
The limit that bites first is usually a monthly cap on messages or resolutions, not the base price. Read these five lines on any plan before you commit.
| What to check | Why it matters | Example from the pages I read today |
|---|---|---|
| Monthly volume cap | Decides when AI replies stop or an extra charge begins | MagizAI pauses at a plan's message cap and does not bill past it; Chatbase Hobby is 700 message credits; Intercom has no message cap and bills $0.99 per outcome |
| Content or training limit | Decides how much of your material the bot can hold | Chatbase Hobby allows 40 MB per agent; MagizAI allows 250 knowledge sources on its lower plans and no limit on Pro and Business |
| Who pays for the model | Decides whether the AI cost sits inside the plan or on your own bill | MagizAI BYOK uses your own provider key; its Starter plan and up include managed keys |
| Removing the vendor badge | A visible "powered by" line, and the cost to drop it | Chatbase charges $1,188 a year to remove its branding; Tidio charges $16.67 a month; MagizAI includes white label on Pro and Business |
| Seats and bots | How many staff and separate sites the plan allows | MagizAI gives 3 seats and 3 bots at entry; Chatbase Hobby gives 2 members and 1 agent |
The three billing models are not comparable by price alone. A flat message cap, as on MagizAI and Chatbase, is predictable but stops when you hit it. A per-outcome charge, as on Intercom Fin and Tidio's Premium pay-per-resolution tier, scales with volume and can cost more as that volume grows. A bring-your-own-key plan moves the model cost onto your own provider account, where you set the spending limit.
What does a custom chatbot cost to build and run?
A custom chatbot has three separate costs: the build, the model usage and the upkeep, each broken down in my guide to what an AI agent or chatbot costs to build. With me, building one that answers from your documents is about 35 to 70 hours of work by my planning estimate, at $55 to $65 an hour or a fixed quote; that is the AI chatbot development route, for needs no product covers.
- Build. Collecting your real questions, cleaning and indexing the content, writing the rules and hand-off, and testing on your questions before launch. Messy PDFs, conflicting policies or a second language push the hours up.
- Running. The model charges per use, billed to your own provider account and not to me. A per-conversation limit, a per-visitor rate limit and a daily spend cap keep it predictable.
- Upkeep. Reading logs, fixing the content behind wrong answers, and retesting when a provider changes a model. A few hours a month, done by your staff or by me.
A subscription tool folds all three into one monthly fee, which is why it wins for common needs. A custom build pays off when the fee for the features you need, or the limits you keep hitting, cost more over a year than the build, or when no product can do the job at all.
What happens to your data and your customers' data?
The visitor's message, the recent conversation and the content chunks needed for the answer go to the model provider's API, and nothing else should. How that data is handled is a question to put to any tool or developer before launch.
- Redaction and logging. Email addresses, phone numbers and card-like numbers should be masked before text is sent or stored. Pattern matching misses some formats, so retention and access limits matter too.
- Retention. Decide how long conversation logs are kept and who can read them, then set it. Chatbase, for example, states on its pricing page that customer data is not used to train models, and lists GDPR, SOC 2 and HIPAA.
- Provider terms. Both OpenAI and Anthropic publish separate data terms for their business APIs; Anthropic's privacy center notes that its consumer and commercial products are governed by different terms. Read the current ones for your plan.
Tell visitors they are talking to an AI, and ask your own adviser what your privacy policy must say about an AI provider processing chat text. No chatbot makes you compliant with any law by itself.
How do you set up and judge a website chatbot before trusting it?
Set it up on a test page first, feed it your own content, and run your real questions through it before a single visitor sees it. The order below keeps a launch from going wrong in public.
- Collect 50 to 100 real questions from tickets, chat logs and sales emails, in the wording customers actually use.
- Give the tool your content: point it at your site, upload your FAQs and PDFs, and remove pages that contradict each other so each fact has one source.
- Add the widget to a staging page or a hidden section, not the live home page.
- Run your question set, read every wrong or weak answer, fix the content or the rule behind it, and run it again.
- Turn on logging and a spend cap, then launch on one page or section and read the real conversations each week.
Checklist before you trust it in public:
- It answers from your content and shows the source page.
- It says it does not know, and offers a person, when nothing relevant is found.
- It passes your own real questions, not a vendor demo.
- Account answers come from the login session, never from a typed email.
- A spend cap and an alert are set, and you know the cost per conversation.
- Logs are masked, kept for a set period, and readable only by named staff.
When is an AI chatbot for a website the wrong choice?
It is the wrong choice when your content cannot support it, or when a wrong answer or action costs more than the bot saves. A few cases where I would wait or choose something else:
- Thin or outdated content. The bot repeats what you give it, so fix the pages first. A good FAQ page may be enough for a dozen stable questions.
- High-stakes answers. Medical, legal or financial advice to the public is not a job for a public bot.
- Actions without a person. Refunds, cancellations and account changes belong behind a confirmation step, not inside an unattended chat.
- A promised resolution rate. Any figure depends on your content and your customers, so measure it on your own site rather than trusting a headline number.
- No way to test first. If you cannot try it on your own questions before paying, that is reason enough to look elsewhere.
What else do people ask about an AI chatbot for a website?
What is the best AI chatbot for a website?
There is no single best one; the right tool depends on your content, your volume, and whether you need account lookups or actions. For answers from your pages and files, a general tool such as Chatbase or Tidio is enough. For hosting and WHMCS, a tool built for that, such as MagizAI, saves setup; the WHMCS AI chatbot page compares the ready-made, workflow and custom routes. Try two on your own questions before deciding.
Can I add an AI chatbot to my website for free?
Partly. Of the tools I checked today, Tidio has a free plan that includes the first 50 Lyro AI conversations for life, while Chatbase and Intercom offer time-limited free trials rather than a free tier. MagizAI has no free tier but a full refund within seven days. A free tier is fine for testing, but watch the monthly limit before you rely on it.
How do I train an AI chatbot on my own website content?
You do not retrain the model; you point the tool at your content and it looks the content up at question time. Most tools let you add your site URL, upload PDFs and DOCX files, and paste question-and-answer pairs. Remove pages that contradict each other first, because the bot cannot tell which one is right.
Do I need to know how to code to add an AI chatbot?
No, for a ready-made tool. Adding one is usually a single line of script before the closing body tag, or a plugin on WordPress. Coding is only needed for a custom build, or to connect the bot to systems a product does not already support.
Is my customers' data safe with an AI chatbot?
It depends on the tool's handling, not on the idea. Check that personal data is masked in logs, that logs have a retention limit and restricted access, and that the model provider's business terms keep your data out of training. Account lookups must use the login session, never an email typed into the chat.
Whichever route fits, I will tell you on a short scoping call which one costs you less, since I sell both a custom build and a hosted product.
Last reviewed 20 October 2026. Disclosure: I sell AI chatbot development and I am founder and CEO of MagizAI, one of the tools compared here, so I am not neutral. I am CTO at ElySpace, a freelance developer for Marmaids LLC, and founder of WHMCSPilot and MagizAI. Prices come from each vendor's pricing page as I read it on that date.
